Single-photon synthetic aperture rapid imaging device and method for gas leakage

Abstract

The invention discloses a single-photon synthetic aperture rapid imaging device and method for gas leakage, and belongs to the technical field of gas detection and imaging. Aiming at the problems in the prior art, on the basis of an SPAD array detector and WMS modulation and demodulation, an FPGA is used for realizing modulation reference synchronization and photon event counting, secondary harmonics of an absorption signal are extracted through digital phase-locked demodulation, and a multi-frame PIC projection drawing is obtained; introducing a small-rotation-angle multi-view-angle acquisition and pose calibration mechanism, and performing synthetic aperture type gas reconstruction by adopting back projection / convolution back projection; two-dimensional FFT is combined with a Gaussian window to realize fast calculation of a convolution kernel frequency domain, and GPU visual angle block parallel and end-to-end data flow organization are adopted to realize fast accumulation reconstruction, so that two-dimensional gas concentration distribution is output, and deep layered gas image display is realized. The device has single-photon-level sensitivity, strong background suppression and rapid imaging capability, can realize rapid positioning and visual imaging of leaked gas plume, and has multi-gas compatibility and platformization expansibility.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of gas detection and imaging, and particularly relates to a single-photon synthetic aperture fast imaging device and method for gas leakage. BACKGROUND

[0002] In application scenarios such as urban gas transmission and distribution networks, petrochemical device areas, storage and transportation stations, and power equipment insulation systems, gas leakage often presents characteristics such as fast diffusion, unstable form, and significant influence from wind fields and obstacles. On-site disposal not only needs to determine "whether there is leakage", but also needs to give spatial information such as "leakage position, diffusion direction and intensity change" within a short time to support rapid judgment and emergency disposal. Therefore, the engineering demand gradually shifts from traditional point detection to remote imaging detection schemes with spatial resolution capability.

[0003] At present, in the aspects of leakage positioning and visualization, the common schemes usually still have the problem of "measurement not matching target demand". Many optical absorption type imaging methods directly obtain a projection quantity (such as path integral form absorption quantity or path integral concentration) along the line of sight. The projection quantity is essentially a superposition of contributions from different spatial positions along the light path, and thus under single-view conditions, the situation of "projection map has response but spatial position is not unique" is prone to occur. Especially in the working conditions where plume has obvious three-dimensional structure, there are obstructions or multi-layer diffusion, it is difficult to stably and accurately give the leakage source position and plume spatial form only by a single frame projection thermal map, resulting in insufficient positioning reliability.

[0004] In addition to the above projection type imaging scheme, infrared thermal imaging and laser absorption spectroscopy are also commonly used for leakage detection in engineering, but there are still limitations in spatial positioning capability and dynamic response. Although the infrared thermal imaging technology is widely used, its imaging effect is easily affected by factors such as background temperature fluctuation, reflection characteristics and gas concentration. In the scene where large-scale leakage or diffusion changes rapidly, the response speed and positioning accuracy are often difficult to balance; especially in low concentration conditions (such as less than 100 ppm), the sensitivity further decreases. Although the laser absorption spectroscopy (such as TDLAS) technology has high detection sensitivity and selectivity, it is essentially point measurement-based, and usually needs to rely on scanning or multi-point arrangement to obtain spatial information, resulting in increased system complexity and increased scanning time. In the scene where the leakage source changes rapidly or the coverage is large, it is difficult to stably meet the dual requirements of real-time and spatial positioning accuracy.

[0005] To improve the spatial positioning capability, the engineering usually introduces scanning or multi-view observation, and the geometric fusion or reconstruction is carried out through the projection information of multiple views. This idea can theoretically convert the projection information into the spatial distribution of gas, but in the process of landing, it is easily restricted by two key factors: on the one hand, multi-view sampling needs to change the view angle in a limited time and keep the pose calibration and time synchronization; on the other hand, multi-view reconstruction usually involves back projection, convolution back projection or other reconstruction operations, and the calculation amount increases rapidly with the pixel resolution, the number of views and (if necessary) the number of depth layers, and the traditional serial processing method cannot meet the requirements of real-time or quasi-real-time output on site.

[0006] In addition, the face array detection and event type data output will further increase the end-to-end processing pressure of the system: when the detection end outputs a high-throughput pixel data stream or a photon event data stream, if the reconstruction end still uses the nested loop method for calculation, not only the calculation redundancy is significant, but also in the case of multi-frame fusion, it is easy to form obvious processing delay, so that the system is difficult to continuously output stable spatial results in the process of dynamic change of leakage. At the same time, the projection graph is easily affected by noise fluctuations and limited field of view in the edge area, causing unclear boundaries or artifacts, which further affects the reliability of the plume profile and positioning results.

[0007] Therefore, a new gas leakage imaging device and method is needed: on the basis of ensuring high-sensitivity projection measurement capability, through small-angle multi-view sampling and pose calibration to enhance spatial constraints, and using synthetic aperture fast reconstruction to realize fast mapping from projection to spatial distribution; at the same time, introducing frequency domain fast convolution and parallel acceleration mechanism to reduce the reconstruction time, so as to realize the fast positioning and two-dimensional / three-dimensional visualization imaging of the leakage gas plume under complex working conditions. SUMMARY

[0008] In view of the problems that the existing gas leakage imaging scheme is difficult to directly obtain the spatial positioning information and three-dimensional shape of the leakage plume, and the calculation amount is significantly increased after introducing multi-view fusion / spatial reconstruction, and it is difficult to meet the requirements of on-site fast response, the present application provides a single-photon synthetic aperture fast imaging device and method for gas leakage. On the basis of maintaining the advantages of single-photon high-sensitivity detection and modulated absorption demodulation, through small-angle multi-view sampling and pose calibration to enhance spatial constraints, and using synthetic aperture back projection / convolution back projection and FFT, GPU block parallel acceleration, the fast two-dimensional / three-dimensional imaging and positioning of the leakage gas plume are realized. The specific principles include the following six aspects:

[0009] 1. Wavelength modulation absorption (WMS) and harmonic shift principle of absorption information

[0010] Gas molecules have selective absorption to laser at characteristic spectral lines. Let the optical frequency corresponding to the laser center wavelength be By modulating the driving to make the laser frequency swing periodically and slightly around the spectral line, it can be expressed as

[0011]

[0012] Or equivalently expressed in wavelength form as

[0013]

[0014] Where (or ) is the modulation depth, is the modulation frequency. Since the absorption coefficient varies with time, the return photon arrival rate appears harmonic components at the modulation frequency and its multiples. Compared with directly measuring the DC component, WMS moves the absorption information from low frequency to and etc. frequency band, so that the system can still extract the absorption characteristics through coherent demodulation under strong background and slow varying noise conditions, and improve the anti-interference ability and sensitivity.

[0015] For the imaging scene based on scattered echo, the effective photon arrival rate at pixel of the receiving end can be expressed as

[0016]

[0017] Where varies with gas absorption modulation, represents the background photon arrival rate. Since the absorption information has coherent characteristics at the harmonic, the subsequent extraction of the harmonic component through digital phase-locked loop can still stably obtain the observation related to the absorption intensity under the condition that statistical noise is dominant.

[0018] 2, Single photon array (SPAD) event statistics and time stamp output principle

[0019] SPAD pixels work in Geiger mode, which can trigger avalanche and produce digital pulse output for a single incident photon, and are naturally suitable for low light and long distance scattered echo scenes. For pixel , its count in a short time window can be approximately regarded as a Poisson process:

[0020]

[0021] Therefore, the system's observation of echoes is essentially an "event-based / count-based" random sequence rather than a continuous analog voltage. The SPAD array can output a full-field-of-view count / event stream in parallel and can optionally provide arrival timestamps. (The i-th event) provides additional constraints for subsequent distance-gated layering and 3D voxel imaging.

[0022] When the system has timestamp capabilities, distance gating can be further introduced: based on the round-trip propagation relationship, the distance corresponding to an event is approximately:

[0023]

[0024] in, The speed is the speed of light. By binning the timestamps (range bins), observations can be expanded from a two-dimensional pixel plane to a data cube of "pixel × distance layer", providing a priori structure for deep layered reconstruction and three-dimensional plume morphology characterization.

[0025] 3. FPGA Synchronization and Modulation Period Counting Sequence Construction Principle

[0026] Since WMS relies on modulation phase information, the random events output by SPAD must be strictly aligned with the modulation reference. This invention utilizes an FPGA to generate the modulation reference and establish a unified time base, completing the mapping between timestamps and modulation phases within the FPGA. Using the modulation period T... m =1 / Using this as the basic unit, the discrete sampling points within the period are numbered as follows: The sampling interval satisfies:

[0027]

[0028] Then, a modulation period counting sequence is constructed:

[0029]

[0030] in, Represents pixels Event counting within the k-th phase sampling interval. This sequence transforms the "asynchronously arriving" event stream into a "phase-synchronized" counting sequence, enabling phase-locked demodulation to perform inner product operations directly in the discrete domain and facilitating pipelined, parallel accumulation, and buffer management on the hardware side. This end-to-end data flow design of "synchronous organization—demodulation output—re-fusion reconstruction" significantly reduces data processing overhead in multi-view scenarios.

[0031] 4. Principle of Digital Phase-Locked Demodulation for Second Harmonic Extraction and PIC Projection Generation

[0032] The modulation period count sequence {N} for each pixel p(k)}, This invention uses the concept of digital phase-locked loop to extract 2 Component. Definition = Then pixel The orthogonal inner product satisfies:

[0033]

[0034]

[0035] The second harmonic amplitude is:

[0036]

[0037] Under the WMS framework The absorption intensity along the optical path corresponding to this pixel has a monotonic relationship and can be mapped to path integral concentration (PIC) through calibration. This invention employs a linear calibration method:

[0038]

[0039] in These are the system calibration coefficients. This processing yields the single-view PIC projection image PIC(u,v), where (u,v) are the pixel plane coordinates. It should be noted that the PIC projection image is an integral quantity along the line of sight. A single-view PIC is usually insufficient to uniquely determine the true spatial location and three-dimensional morphology of the plume; therefore, multi-view constraints and spatial reconstruction mechanisms are necessary.

[0040] 5. Spatial constraint principle of small-angle multi-view sampling and pose calibration

[0041] Recovering spatial distribution from projection requires independent constraints from multiple different observation directions. While large-angle scanning offers strong geometric independence in engineering, it also leads to problems such as long scanning time, heavy mechanical burden, accumulated pose errors, and increased data redundancy, hindering rapid on-site response. This invention employs a "small-angle multi-view" strategy: acquiring M frames of PIC projection within a relatively small angular range and simultaneously recording the pose parameters of each frame. (Azimuth, pitch, displacement / attitude, etc.), forming a data set {PIC^(m)(u, v), }

[0042] Pose calibration is used to determine the projection mapping relationship between a spatial point r ((x, y) in 2D and (x, y, z) in 3D) and pixel coordinates (u, v), denoted as:

[0043]

[0044] Thus, the multi-view projection is unified into the same spatial grid framework. The core of the small angle strategy is to control the number of views, the acquisition time and the processing scale within the range that can be realized in real time while maintaining the gradual enhancement of spatial constraints, so that the system can continuously output stable results in the process of plume rapid change.

[0045] 6. Synthetic aperture de-projection / convolution de-projection fast reconstruction and FFT, GPU block parallel acceleration principle

[0046] The present application characterizes the gas spatial distribution as a grid / voxel function C(r). Based on multi-view PIC projection and pose mapping, synthetic aperture de-projection can be used for fast reconstruction: for each spatial point r, read the corresponding projection value according to different views and weighted accumulation to obtain C(r). In order to compensate for the geometric attenuation, field of view coverage difference and pixel effectiveness under different views, a weight function W m (r) is introduced, and the de-projection form can be expressed as:

[0047]

[0048] In order to improve the stripe artifacts and edge instability problems caused by limited view angle, the present application further adopts convolution de-projection: first, the projection map is convolved (convolution kernel h), and then de-projection accumulation is performed. In order to weaken the edge effect and frequency domain leakage, a Gaussian window function is introduced in the convolution implementation:

[0049]

[0050] The convolution preprocessing uses two-dimensional FFT to convert spatial convolution to frequency domain point multiplication, realizing fast calculation:

[0051]

[0052] The main computational burden of multi-view reconstruction comes from repeatedly performing "projection mapping-interpolation reading-weight calculation-accumulation update" on a large number of spatial grid points under multiple views. The present application introduces an acceleration mechanism of "view block + GPU parallel accumulation": the view index set is divided into a view block every 256 views, and the projection mapping and block accumulation are completed in parallel on the GPU. The intermediate result is directly merged in the GPU memory, avoiding frequent CPU to GPU round-trip transmission; the convolution preprocessing also uses batch processing FFT to realize parallel implementation on the GPU, thereby forming an end-to-end high-speed pipeline.

[0053] When the SPAD array provides timestamps and can perform distance gating, the events can be layered according to distance to obtain layered projection , and the convolution de-projection and parallel accumulation are performed on each layer to obtain a three-dimensional voxel concentration field , so as to maintain the rapidity while obtaining the plume depth information and the spatial positioning ability of the leakage source.

[0054] In summary, on the basis of single-photon WMS projection measurement, the application highlights the introduction of small-angle multi-view constraints, convolution back-projection fast implementation in the frequency domain, Gaussian window suppression of edge effects, and the cooperative principle of GPU view block parallel end-to-end acceleration, which upgrades the leaked plume from the "projection heat map display" to the "two-dimensional / three-dimensional spatial distribution imaging that can be quickly reconstructed", meeting the engineering needs of on-site rapid response.

[0055] In order to achieve the above purpose, the application adopts the following technical scheme:

[0056] A single-photon synthetic aperture fast imaging device for gas leakage, the device comprises the following modules: a laser module, a beam expansion and shaping module, a gas region to be measured, a scattered light acquisition and filtering module, a single-photon detection array module, an FPGA control and synchronization module, a digital phase-locked demodulation module, a gas concentration inversion and image output module, a multi-view acquisition and pose calibration module, a synthetic aperture fast reconstruction module, and a GPU parallel acceleration module; the data flow between each module is: WMS modulation and illumination→SPAD array event / count acquisition→FPGA phase synchronization and sequence organization→pixel-level phase-locked demodulation→PIC projection generation→small-angle multi-view acquisition and pose input→convolution back-projection / back-projection reconstruction→GPU block parallel output of two-dimensional / three-dimensional plume image. Wherein:

[0057] The laser module comprises a tunable DFB laser, a laser temperature control unit, a sinusoidal current modulator, a power supply voltage stabilizing module and a fiber coupler, which is used to output wavelength modulated laser around the target gas absorption spectral line; the instantaneous wavelength of the laser satisfies:

[0058]

[0059] Wherein, is the center wavelength, is the modulation depth, is the modulation frequency. The temperature control unit is used to keep stable and suppress drift, so that the absorption response can be repeatedly extracted between different observation frames.

[0060] The beam expansion and shaping module comprises an aspheric collimating lens, a high-transmittance beam expander lens group, an integrator and a laser output window, which is used to form a surface array irradiation spot covering the gas region to be measured; the integrator adopts a combination of a diffuser sheet and an integrating rod to reduce the risk of detection saturation caused by excessively high central light intensity, while improving the effective number of photons of the edge pixels and improving the spatial consistency of the projection data.

[0061] The to-be-tested gas region includes an open space or a closed gas chamber to be tested and a laser propagation path length control device; the to-be-tested gas region contains a leakage gas plume mixed area with background gas; the device can be optionally provided with a path length control member or a mirror frame to limit or change the propagation path of the laser in the scene, so as to facilitate the formation of a stable measurement geometric relationship at different distances and in different observation directions, and improve the consistency of calibration and reconstruction.

[0062] The scattered light collection and filtering module includes a high-flux mirror group or a Fresnel lens, a central wavelength band-pass filter, an optical black cavity structure and a light shield, which are used for collecting backward or near-backward scattered photons of the to-be-tested gas region and suppressing non-target waveband background light; preferably, the central wavelength of the filter is matched with The bandwidth is set to ±(1-5) nm to balance the passing of gas absorption signals and the suppression of solar background; the light-shielded black cavity is used for reducing stray light into the detection surface array.

[0063] The single-photon detection array module includes a CMOS-compatible SPAD array chip, an on-chip avalanche current trigger identification circuit, a per-pixel independent time-to-digital converter or a full-array shared TDC module and a light window cover plate, which are used for counting incident photon events per pixel and outputting pixel positions and photon arrival time stamps; the TDC provides nanosecond-level or higher resolution time marking, so that the device can obtain pixel counts and time-gated count sequences of different distance segments, providing a data basis for depth-layered reconstruction.

[0064] The FPGA control and synchronization module includes an FPGA chip, a DDS modulation signal generator, a laser modulation control pin, a SPAD time marking alignment logic, a multi-channel FIFO buffer, a DMA transmission interface and a high-stability clock source, which are used for outputting a laser modulation reference signal and aligning the photon event time stamps with the modulation signal phase, and organizing the photons of each pixel in each modulation period into a count sequence;

[0065] The digital phase-locked demodulation module includes a digital multiplier, an accumulator array, a sine / cosine wave lookup table, a modulation phase alignment controller, a second harmonic amplitude calculation module and a pixel demodulation matrix register group, which are used for inner product processing of the count sequence and a reference harmonic signal and outputting a pixel-level second harmonic amplitude . The inner product of the pixel count sequence and the orthogonal reference sequence is:

[0066]

[0067]

[0068]

[0069] in, This represents the number of sampling points within one modulation period. This demodulation method is suitable for SPAD event-type data and can selectively extract absorbed signals under strong background light conditions. This represents the in-phase integral of the pixel count sequence and the second harmonic cosine reference signal; This represents the orthogonal integral of the pixel count sequence and the second harmonic sinusoidal reference signal. Indicates the integral quantity within the in-phase phase. Integrals within orthogonal regions The synthesized pixel-level second harmonic amplitude is used to characterize the gas absorption intensity in the optical path corresponding to that pixel.

[0070] The gas concentration inversion and image output module includes a path integral concentration calculator, a pixel normalization processing unit, a heatmap encoding and pseudo-color mapping module, an image frame buffer, and a display / network output interface, used to convert pixel-level second harmonic amplitude into a path integral concentration PIC projection image and output it; the PIC satisfies:

[0071]

[0072] in, This module provides a system calibration factor obtained from a standard gas chamber or a known concentration path. It can also perform normalization, bad spot repair, dynamic range compression, and threshold enhancement to generate stable projection data that can be used for subsequent synthetic aperture reconstruction.

[0073] The multi-view acquisition and pose calibration module includes a turntable / pan-tilt scanning mechanism or a galvanometer scanning mechanism and a pose parameter recording unit, which is used to acquire multiple frames of PIC projection images within a small turning angle range and record the azimuth angle, pitch angle and / or pose parameters corresponding to each frame; the pose parameters can be obtained by the angle encoder of the turntable, pan-tilt or scanning mechanism, or recorded by an external pose sensor;

[0074] The synthetic aperture rapid reconstruction module is used to perform back-projection reconstruction or convolutional back-projection reconstruction on multi-view PIC projection images based on the pose parameters to obtain two-dimensional gas distribution maps, deep layered gas images or three-dimensional voxel concentration fields.

[0075] The synthetic aperture rapid reconstruction module is used to invert multi-view PIC projection images into two-dimensional gas distribution maps, depth-layered gas images, or three-dimensional voxel concentration fields based on the pose parameters. This module includes at least: a coordinate mapping unit, a convolution kernel generation unit, a 2D FFT convolution unit, and a back-projection accumulation unit. For two-dimensional planar reconstruction, let the plane to be reconstructed be... The grid, the projection of the i-th viewpoint The module first... Multiply by Gaussian window to suppress the edge ringing caused by limited field of view; Gaussian window satisfies

[0076]

[0077] convolve the windowed projection with the corresponding geometric propagation kernel to get ; finally backfill and accumulate the result according to the geometric mapping to get the reconstruction result:

[0078]

[0079] To further reduce the computation of convolution, the 2D FFT convolution unit realizes the above-mentioned convolution in frequency domain:

[0080]

[0081]

[0082] Compared with point-by-point spatial convolution, the frequency domain implementation reduces the computation complexity from to , and is naturally suitable for GPU parallel implementation, thereby supporting real-time or quasi-real-time gas leakage imaging.

[0083] The GPU parallel acceleration module is used for dividing the convolution operation and the back projection accumulation operation of the convolution back projection reconstruction into view angle blocks and performing parallel accumulation output on the GPU end, so as to realize fast visualized imaging of gas leakage plume.

[0084] The frequency of the DDS modulation signal generator is set to 10 kHz, and the frequency of the high-stability clock source is ≥100 MHz; the small rotation angle range of the multi-view acquisition and pose calibration module is 5°-45°, preferably ±5°-±10° around the reference azimuth center angle; when performing convolution back projection reconstruction, the synthetic aperture fast reconstruction module introduces a Gaussian window function to the projection image or the convolution kernel to weaken the edge effect, and the Gaussian window function satisfies:

[0085]

[0086] wherein, , is the projection plane coordinate, is a window width parameter, and the size is generally consistent with the observation range;

[0087] ​The GPU parallel acceleration module divides 64-1024 view angles in a view angle set into a block, preferably 256 view angles into a block according to the size of an actual GPU memory, and completes in-block accumulation and inter-block merging at the GPU end; when the SPAD array chip outputs a photon arrival time stamp, the synthetic aperture fast reconstruction module further performs hierarchical reconstruction based on distance gating to output a depth hierarchical gas image or a three-dimensional voxel concentration field.

[0088] A single-photon synthetic aperture fast imaging method for gas leakage, the method comprising the following steps:

[0089] Step 1: A continuous wave laser with a center wavelength near the absorption spectrum of the target gas is emitted by a laser module, and a sinusoidal current signal is superimposed by a FPGA control and synchronization module to form a wavelength modulation laser by slightly scanning the laser wavelength near the absorption line; the formula of the instantaneous emission wavelength is:

[0090]

[0091] Wherein, λ(t) represents the instantaneous emission wavelength at time t, λ0 represents the center wavelength, Δλ represents the wavelength modulation depth, f m represents the modulation frequency;

[0092] Step 2: The modulation laser is spatially enlarged and light intensity homogenized by a aspheric collimating lens and a high-transmittance beam expander lens group to form a planar spot to irradiate the target scene covered by the gas to be measured;

[0093] Step 3: When the modulation laser passes through the gas to be measured, selective absorption occurs near the characteristic wavelength of the target gas molecules, and scattered photons in the backward or near-backward direction are generated;

[0094] Step 4: The scattered photons are collected by a scattered light collection and filtering module, and after background interference is suppressed by a center wavelength band-pass filter, they are sent to a SPAD array for pixel-by-pixel photon event recording, and the pixel position and photon arrival time stamp are recorded;

[0095] Step 5: The FPGA control and synchronization module outputs a laser modulation reference signal and records the system clock, aligns the time stamp of all photon events with the phase of the modulation signal, realizes periodic photon statistics, and organizes the photon array of each pixel in each modulation period into a count sequence in each modulation period.

[0096] Step 6: The count sequence of each pixel is processed by orthogonal inner product with a second harmonic reference signal to extract the second harmonic amplitude value; the second harmonic amplitude value satisfies:

[0097]

[0098]

[0099]

[0100] wherein, represents the count sequence of a pixel at discrete time k is the number of sampling points in a modulation period, , is the sampling time interval, is the second harmonic amplitude of the pixel; I represents the in-phase inner product component of the pixel count sequence and the second harmonic cosine reference signal; Q represents the quadrature inner product component of the pixel count sequence and the second harmonic sine reference signal.

[0101] Step 7: multiply the second harmonic amplitude value of each pixel by the system calibration factor to obtain the path integral concentration PIC projection image; the formula is:

[0102]

[0103] wherein, is the system calibration factor; is the second harmonic amplitude of the pixel.

[0104] Step 8: acquire multiple frames of PIC projection images within a small rotation angle range by the multi-view acquisition and pose calibration module, and record the corresponding azimuth angle, pitch angle and / or pose parameters of each frame to form a multi-view PIC projection set. Since the single-view path integral concentration projection cannot uniquely determine the real distribution position of the gas in space, it is necessary to combine the multi-view projection information and perform back projection or convolution back projection reconstruction according to the pose relationship to obtain the gas distribution result with spatial significance.

[0105] Step 9: perform back projection or convolution back projection reconstruction on the multi-view PIC projection set according to the pose parameters to obtain the gas distribution on the spatial grid; wherein the back projection reconstruction satisfies:

[0106]

[0107] wherein, is the spatial grid point or voxel coordinate, is the pose parameter of the frame, is the projection mapping operator determined by the pose, is the weight function, is the number of views; when convolution back projection is used, each frame of projection image is first preprocessed by convolution and fast convolution is realized by two-dimensional FFT, which satisfies:

[0108]

[0109] where F(·) and F-1(·) are two-dimensional fast Fourier transform and inverse transform respectively, h is a convolution kernel, is a Gaussian window function. represents the path integral concentration projection image after Gaussian window weighting and convolution preprocessing under the th view angle.

[0110] Step 10: The GPU parallel acceleration module divides the view angle index into several view angle blocks, and performs convolution and back projection accumulation in step 9 in parallel. After block accumulation and block merging are completed on the GPU side, the two-dimensional gas distribution map is output at one time, which is used for fast positioning and visual imaging of the leaked gas.

[0111] Compared with the prior art, the present application has the following advantages:

[0112] 1. Based on single-photon wavelength modulation absorption projection measurement, the present application introduces small-angle multi-view acquisition and pose calibration, and fuses and reconstructs multiple projection information through synthetic aperture inversion, so that the output result is no longer limited to the two-dimensional projection display of path integral concentration, but forms a two-dimensional gas distribution imaging result with spatial significance, thereby more intuitively representing the leakage source position and gas plume diffusion shape, and supporting qualitative and quantitative evaluation of the leakage range and intensity change.

[0113] 2. In view of the large calculation amount and low processing efficiency of the traditional inversion method in multi-view imaging, the present application organizes the key operation in the inversion process into a form suitable for batch processing to effectively reduce repeated calculation and redundant operation, significantly improves the overall processing efficiency while ensuring the imaging quality, and is more suitable for fast imaging requirements under large field of view and high resolution conditions.

[0114] 3. Due to the limitation of the field of view range and noise condition, the traditional reconstruction result is prone to stripe artifacts or contour discontinuity in the edge region. The present application performs window preprocessing on the projection data before reconstruction, which helps to weaken the influence of edge effect and frequency domain leakage, improves the continuity and stability of the reconstruction result in the boundary region, and enhances the readability and interpretation reliability of the gas plume contour.

[0115] 4. The present application introduces the view angle block idea in the multi-view reconstruction process, organizes and processes multiple projection data in blocks in parallel, effectively reduces the delay and resource occupation caused by serial calculation, reduces the data transmission and storage pressure, thereby improving the end-to-end processing efficiency of the system, and meeting the fast imaging requirements under the condition of multi-view and multi-frame fusion.

[0116] ​5. The application adapts the event-type or count-type data output by the single-photon detection array, obtains stable projection measurement results through unified data synchronization and demodulation processing, and in the case of having arrival time information, can further combine distance gating to realize layered imaging or spatial distribution reconstruction, and in the case of not having time information, can still realize two-dimensional gas distribution imaging, thereby improving the applicability of the device under different hardware conditions.

[0117] 6. The application combines modulated absorption demodulation and multi-view fusion reconstruction, and in the case of strong background light, complex scattering conditions or decreased quality of part of the views, can still output relatively stable gas distribution imaging results, reduces false positives and artifacts, and improves the reliability of the leakage detection and positioning results.

[0118] 7. The application adapts the event-type or count-type data output by the single-photon detection array, obtains stable projection measurement results through unified data synchronization and demodulation processing, and in the case of having arrival time information, can further combine distance gating to realize layered imaging or spatial distribution reconstruction, and in the case of not having time information, can still realize two-dimensional gas distribution imaging, thereby improving the applicability of the device under different hardware conditions.

[0119] 8. The application has clear overall structure and clear module division, the laser and the optical assembly can be configured and replaced according to different target gas absorption spectral lines, can be applied to fixed monitoring scenes, can be extended to mobile platforms such as vehicles, portable devices or unmanned aerial vehicles, and is suitable for various gas leakage detection and visualization imaging application scenes such as urban gas, petrochemical device area and power equipment. BRIEF DESCRIPTION OF DRAWINGS

[0120] Figure 1 It is a device diagram of the single-photon gas detection rapid imaging device of the overall application device;

[0121] Figure 2 It is a system optical path structure schematic diagram of the application device, which shows the process of laser emission, beam expansion and light homogenization, scene irradiation, gas absorption and backscattered light collection;

[0122] Figure 3 It is a SPAD array detection and synchronization control structure schematic diagram, which shows TDC time marking, FPGA phase alignment, data buffering and output interface;

[0123] Figure 4 It is a functional module composition block diagram of the application device, which includes a laser module, an optical module, a SPAD detection array, an FPGA control and demodulation module, a reconstruction and display module;

[0124] Figure 5 It is a small-angle multi-view acquisition and pose calibration schematic diagram, which shows the process of acquiring multiple frames of projections and recording pose parameters within a limited angle range.

[0125] Figure 6 This diagram illustrates the process of rapidly implementing convolutional backprojection and 2D FFT, showing the processing chain of windowing, FFT, frequency domain multiplication, IFFT, and backprojection accumulation.

[0126] Figure 7 The images show a blurred two-dimensional gas plume at a single angle, and two-dimensional gas plume imaging at multiple angles at small angles.

[0127] Figure 8 The diagram shows the overall workflow of the method of this invention, labeled sequentially as: wavelength modulation → beam illumination → SPAD photon detection → harmonic demodulation → PIC projection → multi-view acquisition → rapid reconstruction → imaging display. Detailed Implementation

[0128] To gain a deeper understanding of this invention, we will provide a comprehensive and detailed description. However, this invention has various implementations and is not limited to the specific examples listed herein. These examples are presented to enhance a full understanding of the disclosure of this invention.

[0129] like Figures 1-6 As shown, this embodiment presents a method for achieving rapid two-dimensional imaging without relying on precise arrival timestamps, applicable to SPAD area array detectors that only output pixel counts or only output period count sequences. Furthermore, in the absence of photon arrival time information, this invention preferably implements two-dimensional gas distribution imaging based on multi-view constraints; when the single-photon detection array can output photon arrival timestamps, range gating can be further combined to achieve layered imaging or three-dimensional voxel reconstruction. The device is arranged as follows... Figure 3 It consists of a laser module, a beam spreading and shaping module, a scattered light acquisition and filtering module, a SPAD array module, an FPGA control and synchronization module, a digital phase-locked demodulation module, a synthetic aperture fast reconstruction module, and a GPU parallel acceleration module.

[0130] (1) Laser modulation and illumination parameters: The center wavelength is selected near the characteristic absorption line of the target gas (taking CH4 as an example). (e.g., 1.65 μm band), modulation uses sinusoidal current modulation, modulation frequency... Preferably 1–50 kHz, more preferably 5–20 kHz; modulation depth The laser beam is set according to the absorption linewidth and the desired second harmonic amplitude. After beam expansion and homogenization, the laser beam forms an array of irradiation spots covering the leakage area, with the spot diameter preferably being 50–200 mm.

[0131] (2) Single photon collection and phase synchronization: SPAD array outputs photon events or counts pixel by pixel, and FPGA outputs modulated reference and establishes unified time base. FPGA counts events of each pixel as a period count sequence c according to modulation period p [n] (n = 0…N-1), and completes phase alignment and data caching.

[0132] (3) Second harmonic phase-locked demodulation and PIC projection: orthogonal projection is performed on each pixel to obtain , and the second harmonic amplitude is calculated, and then the path integral concentration projection is obtained by a calibration factor . The projection reflects the integrated concentration along the line-of-sight direction (unit: ppm·m), and a thermal map can be output for rapid warning.

[0133] (4) Small angle of rotation multi-view collection: as shown in Figure 5 , M frames of projections are obtained within a small angle of rotation range by a gimbal or a rotating table, and pose parameters are recorded. M is preferably 11-101 frames, and more preferably 21-61 frames; the total angle of rotation range is preferably ≤10°, and the interval between adjacent viewing angles is preferably 0.05°-1°. To reduce the influence of plume changes, "short-time burst collection" can be used to complete a set of viewing angle collection within 0.2-2 s.

[0134] (5) Convolution back-projection fast reconstruction: as shown in Figure 6 , each frame of projection is windowed, and the window width σ is preferably proportional to the field of view size (for example, σ ≈ 0.2-0.4 times the field of view radius), and then convolution is completed in the frequency domain and back-projection is accumulated. Set the imaging range and resolution for a two-dimensional plane grid (x, y), for example, the imaging range is 0.5-20 m², and the grid size is 256×256 or 512×512; and the reconstruction result Ĉ(x, y) is obtained by taking values according to the pose mapping and accumulating for each grid point.

[0135] (6) GPU block parallelism: the viewing angle set is divided into blocks, each block containing 5-20 frames of viewing angles, and the GPU end completes FFT, frequency domain point multiplication, IFFT, and back-projection accumulation for the block. The block strategy can stably run in the case of limited display memory, and significantly improve the throughput. The output end can superimpose the two-dimensional reconstruction result as a pseudo-color concentration map on the visible light image or scene map to realize intuitive labeling of the leakage position.

[0136] as Figure 3 and Figure 7As shown, the embodiment further utilizes the photon arrival time stamp provided by TDC to realize distance gated layering and 3D voxel reconstruction based on embodiment 1. The device hardware is the same as embodiment 1, the difference is that the SPAD array output contains the arrival time stamp of each photon event , and FPGA or host computer bins it according to distance gate.

[0137] (1) Distance gating and layering projection formation: convert the time stamp to distance according to R = c·t / 2, and set the distance gate width ΔR (preferably 0.05-1.0 m, more preferably 0.1-0.5 m) and the number of gates D (preferably 4-64). Count the cycle count for each distance gate d and demodulate to get , generate layering PIC projection .

[0138] (2) Layering convolution back projection: independently perform windowing, FFT convolution and back projection accumulation for each distance gate d to get layering plane concentration map . Stack the layering results according to the distance gate center R d or depth to form voxel concentration field . When the imaging geometry is near-axis backscattering, the can be approximately mapped to the distance direction.

[0139] (3) 3D display and quantitative output: perform volume rendering or isosurface extraction on the 3D voxel data, output plume 3D morphology, leakage source 3D coordinate estimation and volume / total amount indicators. Optionally, perform time series filtering or Kalman fusion on the 3D results output at each time to get more stable dynamic 3D display.

[0140] In the case of strong background light, occlusion or scanning speed fluctuation leading to the decline of part of the perspective projection quality, this embodiment introduces perspective quality evaluation and weighted fusion mechanism to improve robustness. Specifically, for the i-th frame projection, calculate the average photon number , second harmonic signal-to-noise ratio and background fluctuation index , and construct perspective weight , for example:

[0141]

[0142] where α, β are empirical parameters (preferably α = 0.5-2, β = 0.1-1), and the weight is normalized to make =1. In the back projection accumulation stage, use the weighted form . This way can suppress the contribution of low signal-to-noise ratio perspectives to the reconstruction stripes, and improve the continuity and false alarm robustness of the leakage boundary.

[0143] For the convenience of the skilled in the art, a set of reproducible typical parameter ranges (only as an example, not as a limitation) is given in this embodiment: (a) modulation frequency =10 kHz, N=256 sampling points per cycle; (b) number of views M=41, total angular range ±8°, angular step 0.4°; (c) Gaussian window sigma taken as the projection width; (d) 2D grid 512x512, imaging area 1 m x 1 m to 10 m x 10 m configurable; (e) when timestamp is enabled, ΔR=0.2 m, number of gates D=16; (f) parallel computation using GPU or multi-core processor to implement view binning, 10 frames per bin, using single-precision complex FFT to improve throughput. The output forms include 2D concentration map, layered depth slice map.

[0144] The contents not described in detail in the specification of the present application belong to the prior art known to the skilled in the art. Although the above describes the specific embodiments of the present application for the purpose of facilitating the understanding of the present application by the skilled in the art, it should be clear that the present application is not limited to the scope of the specific embodiments, and for the skilled in the art, it is obvious that all kinds of changes within the spirit and scope of the present application defined and determined by the appended claims are obvious, and all kinds of inventions using the concept of the present application are included in the protection.

Claims

1. A single-photon synthetic aperture fast imaging device for gas leak, characterized in that, The device comprises the following modules: a laser module, a beam expansion and shaping module, a gas region to be measured, a scattered light collection and filtering module, a single-photon detection array module, an FPGA control and synchronization module, a digital phase-locked demodulation module, a gas concentration inversion and image output module, a multi-view acquisition and pose calibration module, a synthetic aperture fast reconstruction module, and a GPU parallel acceleration module; wherein the connection relationship and data flow relationship between the modules are as follows: The output end of the laser module is optically connected to the input end of the beam expansion and shaping module, and the beam expansion and shaping module is used to expand and shape the wavelength-modulated laser output by the laser module and then irradiate it to the gas region to be measured; The gas region to be measured has an absorption and scattering effect on the wavelength-modulated laser, forming scattered light containing gas absorption information; The input end of the scattered light collection and filtering module is directed towards the gas region to be measured, and is used to collect scattered light and perform spectral filtering, and then guide the filtered scattered light to the single-photon detection array module; The single-photon detection array module is electrically connected to the FPGA control and synchronization module, and is used to convert the received scattered light into a photon count or photon event signal and output it to the FPGA control and synchronization module; The FPGA control and synchronization module is connected to the laser module and the single-photon detection array module, and is used to generate a laser wavelength modulation synchronization signal, time-synchronize and periodically count the photon signal output by the single-photon detection array module, and send the synchronized count sequence to the digital phase-locked demodulation module; The digital phase-locked demodulation module is connected to the FPGA control and synchronization module and the gas concentration inversion and image output module, and is used to perform harmonic phase-locked demodulation on the count sequence, extract the modulation harmonic component related to gas absorption, and generate path integral concentration projection data; The gas concentration inversion and image output module is connected to the multi-view acquisition and pose calibration module, and is used to receive path integral concentration projection data under multi-view and corresponding pose parameters; The synthetic aperture fast reconstruction module is connected to the gas concentration inversion and image output module, the multi-view acquisition and pose calibration module, and the GPU parallel acceleration module, and is used to perform synthetic aperture back-projection or convolution back-projection reconstruction based on multi-view path integral concentration projection data and pose parameters, to obtain a two-dimensional gas distribution image, or a three-dimensional gas distribution image under the condition of having distance resolution information; The GPU parallel acceleration module is used to perform parallel acceleration processing on the convolution operation and back-projection accumulation operation in the synthetic aperture fast reconstruction module, to improve the reconstruction speed of gas leakage imaging.

2. The single-photon synthetic aperture fast imaging device for gas leak according to claim 1, wherein: The laser module comprises a tunable DFB laser, a laser temperature control unit, a sinusoidal current modulator, a power supply voltage stabilizing module, and a fiber coupler, and is used to output wavelength-modulated laser around a target gas absorption spectral line; The beam expansion and shaping module comprises an aspheric collimating lens, a high-transmittance beam expander, a homogenizer, and a laser output window, and is used to form a surface array irradiation spot of wavelength-modulated laser covering the gas region to be measured; The to-be-tested gas region comprises a to-be-tested open space or a closed gas chamber and a laser propagation path length control device; the to-be-tested gas region comprises a leakage gas plume mixed area with background gas; The scattered light collection and filtering module comprises a high-flux reflector group or a Fresnel lens, a central wavelength band-pass filter, an optical black cavity structure and a light shield, which are used for collecting backward or near-backward scattered photons of the to-be-tested gas region and suppressing non-target waveband background light; The single-photon detection array module comprises a CMOS compatible SPAD array chip, an on-chip avalanche current trigger identification circuit, a per-pixel independent time-to-digital converter or a full-array shared TDC module and a light window cover plate, which are used for counting incident photon events per pixel and outputting pixel positions and photon arrival time stamps; The FPGA control and synchronization module comprises an FPGA chip, a DDS modulation signal generator, laser modulation control pins, SPAD time stamp alignment logic, multi-channel FIFO buffers, a DMA transmission interface and a high-stability clock source, which are used for outputting laser modulation reference signals and aligning photon event time stamps with modulation signal phases, and organizing photon arrays of each pixel in each modulation period into a count sequence; The digital phase-locked demodulation module comprises a digital multiplier, an accumulator array, a sine / cosine wave lookup table, a modulation phase alignment controller, a second harmonic amplitude calculation module and a pixel demodulation matrix register group, which are used for performing inner product processing on the count sequence and a reference harmonic signal and outputting pixel-level second harmonic amplitudes; The gas concentration inversion and image output module comprises a path integral concentration calculator, a pixel normalization processing unit, a heat map encoding and pseudo-color mapping module, an image frame buffer and a display / network output interface, which are used for converting pixel-level second harmonic amplitudes into path integral concentration (PIC) projection images and outputting the same; The multi-view acquisition and pose calibration module comprises a turntable / pan-tilt scanning mechanism or a galvanometer scanning mechanism and a pose parameter recording unit, which are used for acquiring multiple PIC projection images in a small rotation angle range and recording azimuth angles, elevation angles and / or pose parameters corresponding to each frame; The synthetic aperture fast reconstruction module is used for performing back-projection reconstruction or convolution back-projection reconstruction on the multi-view PIC projection images according to the pose parameters, so as to obtain a two-dimensional gas distribution map, a depth-laminated gas image or a three-dimensional voxel concentration field. The synthetic aperture fast reconstruction module is used for inverting the multi-view PIC projection images into a two-dimensional gas distribution map, a depth-laminated gas image or a three-dimensional voxel concentration field according to the pose parameters.

3. The single-photon synthetic aperture fast imaging device for gas leak facing according to claim 2, characterized in that: The frequency of the DDS modulation signal generator is set to 10 kHz, and the frequency of the high-stability clock source is ≥100 MHz; the small rotation angle range of the multi-view acquisition and pose calibration module is 5°-45°, and the synthetic aperture fast reconstruction module introduces a Gaussian window function to weaken edge effects on the projection image or a convolution kernel when performing convolution back-projection reconstruction, and the Gaussian window function satisfies: wherein, , is a projection plane coordinate, is a window width parameter; The GPU parallel acceleration module divides the perspective set into 64-1024 perspective blocks, and completes the in-block accumulation and inter-block merging at the GPU end; when the SPAD array chip outputs the photon arrival time stamp, the synthetic aperture rapid reconstruction module further performs hierarchical reconstruction based on distance gating to output a depth hierarchical gas image or a three-dimensional voxel concentration field.

4. A method for single-photon synthetic aperture fast imaging oriented to gas leakage, the method is implemented based on the device of any one of claims 1-3, characterized in that, The method comprises the following steps: Step 1: a laser module is used to emit continuous wave laser with a central wavelength near the absorption spectrum line of the target gas, and a sinusoidal current signal is superimposed through an FPGA control and synchronization module to form a wavelength modulation laser by slightly scanning the laser wavelength near the absorption line; the formula of the instantaneous emission wavelength is: where λ(t) represents the instantaneous emission wavelength at time t, λ0represents the center wavelength, Δλ represents the wavelength modulation depth, f m represents the modulation frequency; Step 2: the modulation laser is spatially magnified and light intensity homogenized by an aspheric collimating lens and a high-transmittance beam expander to form a planar spot to irradiate the target scene covered by the gas to be measured; Step 3: when the modulation laser passes through the gas to be measured, selective absorption occurs near the characteristic wavelength of the target gas molecules, and back or near-back scattered photons are generated; Step 4: the scattered photons are collected by a scattered light collection and filtering module, and after background interference is suppressed by a center wavelength band-pass filter, they are sent into a SPAD array for pixel-by-pixel photon event recording, and the pixel position and photon arrival time stamp are recorded; Step 5: an FPGA control and synchronization module outputs a laser modulation reference signal and records a system clock, aligns the time stamp of all photon events with the phase of the modulation signal, realizes periodic photon statistics, and organizes the photon array of each pixel in each modulation period into a count sequence in the modulation period; Step 6: the count sequence of each pixel is subjected to orthogonal inner product processing with a second harmonic reference signal to extract the second harmonic amplitude value; the second harmonic amplitude value satisfies: wherein, represents the count sequence of a pixel at discrete time instants , K represents the number of sampling points within one modulation period, = 2πf m , Δt represents the sampling time interval, represents the second harmonic amplitude of the pixel; I represents the in-phase inner product component of the pixel count sequence with the second harmonic cosine reference signal; Q represents the quadrature inner product component of the pixel count sequence with the second harmonic sine reference signal; Step 7: the second harmonic amplitude value of each pixel is multiplied by a system calibration factor to obtain a path integral concentration (PIC) projection image; the formula is: wherein, represents a system calibration factor; is the second harmonic amplitude of the pixel; Step 8: a multi-view acquisition and pose calibration module acquires multiple PIC projection images in a small rotation angle range, and records the azimuth angle, pitch angle and / or pose parameters corresponding to each frame to form a multi-view PIC projection set; Step 9: according to the pose parameters, the multi-view PIC projection set is subjected to back projection or convolution back projection reconstruction to obtain the gas distribution on the spatial grid; wherein the back projection reconstruction satisfies: wherein, denotes a spatial grid point or voxel coordinate, denotes the pose parameter of the frame, denotes a projection mapping operator determined by the pose, denotes a weight function, denotes the number of views; when using convolution back-projection, a convolution preprocessing is first performed on each frame of projection image and fast convolution is realized through two-dimensional FFT, satisfying: wherein F(·) and F-1(·) denote two-dimensional fast Fourier transform and two-dimensional inverse transform, respectively, h denotes a convolution kernel, (·) respectively denote two-dimensional fast Fourier transform and two-dimensional inverse transform, respectively, h denotes a convolution kernel, denotes a Gaussian window function; denotes the path integral concentration projection image after Gaussian window weighting and convolution preprocessing under the first view angle. Step 10: a GPU parallel acceleration module divides the perspective index into several perspective blocks, and performs convolution and back projection accumulation of step 9 in parallel, and outputs the two-dimensional gas distribution map for one-time output after completing the in-block accumulation and inter-block merging at the GPU end, which is used for rapid positioning and visual imaging of the leaked gas.

Images